Documentation/Agent Builder
Agent Builder

Agent Builder

Build and configure AI agents without writing code. Choose a model, write a system prompt, attach a knowledge base, and deploy — all from the dashboard in minutes.

Creating your first agent

1

Go to Agents → Create New Agent

From the dashboard sidebar, click Agents, then click the Create Agent button in the top right.

2

Choose a template or start blank

Select one of the 9 built-in templates to start with a pre-configured system prompt, or choose General for a blank slate.

3

Set a name, pick a model

Give the agent a name your customers will see. Select your preferred LLM provider and model from the dropdown — you can switch models at any time.

4

Write or refine the system prompt

The system prompt defines the agent's role, tone, and boundaries. Be specific: tell it what it can and cannot answer, and how to handle escalations.

5

Attach a knowledge base

Add your documents, website, or Q&A entries so the agent can give accurate, source-backed answers.

6

Test in the playground

Use the built-in test panel to simulate customer conversations before going live. Adjust the prompt and knowledge base until responses are accurate.

Configuration reference

Identity & persona

  • Agent name and description — visible to your customers
  • Avatar image — upload a logo or choose an icon
  • System prompt — the core instruction set defining the agent's role and limits
  • Tone presets: Professional, Friendly, Empathetic, Technical, or Custom

Model & response settings

  • LLM provider and model selection — choose from 10+ providers
  • Temperature (0–1): lower = consistent and precise, higher = creative and varied
  • Maximum response length: set a token limit to keep answers concise
  • Streaming responses: enable real-time token streaming for a faster feel

Knowledge & context

  • Attach one or more knowledge bases — the agent searches them on every message
  • Context window size: configure how much conversation history the model receives
  • System prompt variables: inject dynamic values like the customer's name at runtime

Testing & iteration

  • Built-in test playground: send messages and see responses without going live
  • Agent comparison: run two agents side-by-side to compare prompts or models
  • Token usage display: see exactly how many tokens each response uses
  • Debug mode: inspect which knowledge chunks were retrieved for each answer

Supported models

SolveoAI routes requests through a unified LLM gateway. Add provider API keys once under Settings → AI Providers, then use them across any agent.

Anthropic

  • Claude 3.5 Sonnet
  • Claude 3.5 Haiku
  • Claude 3 Opus

OpenAI

  • GPT-4o
  • GPT-4o mini
  • GPT-4 Turbo

Google

  • Gemini 1.5 Pro
  • Gemini 1.5 Flash
  • Gemini 2.0 Flash

DeepSeek

  • DeepSeek V3
  • DeepSeek V2.5

xAI

  • Grok-2
  • Grok-2 mini

Mistral

  • Mistral Large
  • Mistral Small
  • Mixtral 8x7B

Groq

  • LLaMA 3.1 70B
  • LLaMA 3.1 8B

Cohere

  • Command R+
  • Command R

Perplexity

  • Sonar Large
  • Sonar Small

Agent templates

Templates give you a pre-configured system prompt and recommended settings as a starting point. Customize everything after selecting one.

Customer Support

Handles product questions, returns, and order inquiries

Sales Agent

Qualifies leads, answers pricing questions, books demos

E-commerce

Product search, cart help, order tracking for online stores

Healthcare

Appointment scheduling, clinic info, health FAQs — not medical advice

IT Helpdesk

Troubleshooting, ticket creation, and internal knowledge base access

Education

Admissions, enrollment, course info, and student support

Fitness & Wellness

Class bookings, membership questions, and trainer FAQs

Travel

Booking support, itinerary questions, and cancellation policies

General

Blank slate — configure for any use case from scratch

A/B testing agents

Compare two agent configurations by creating a second agent with a different model or system prompt and routing a portion of traffic to each. Use the analytics dashboard to compare resolution rate, response time, and sentiment to determine the better setup.

Create two agents

Same knowledge base, different model or prompt

Route traffic

Split visitors using your deployment settings

Compare metrics

Resolution rate, CSAT, and response time side-by-side

Writing effective system prompts

Be explicit about the agent's role

Start with a clear identity: "You are a customer support agent for Acme Inc. Your role is to help customers with product questions, order status, and returns."

Define what the agent should NOT do

Explicitly list off-limits topics: "Do not answer questions about competitors, give financial advice, or discuss topics unrelated to our products."

Specify escalation behavior

Tell the agent when to hand off: "If the customer is frustrated or asks for a refund over $100, say: 'Let me connect you with a specialist' and stop responding."

Set the tone

Describe the desired communication style: formal vs. casual, concise vs. detailed. Give an example of the ideal response style.

Use the knowledge base, not the prompt, for facts

Do not put product catalog information, pricing tables, or policy text directly into the system prompt. Put that in the knowledge base and let retrieval handle it.

Ready to build your agent?

Create your first agent in under 5 minutes — no code required.